AI-Driven Computer Vision for Intelligent Home Automation and Surveillance Systems

This chapter delves into the innovative application of depth map analysis for enhancing the interpretability of artificial intelligence (AI) decision-making processes. The increasing complexity of AI algorithms, particularly deep learning ones, frequently results in a “black-box” situation, where the rationale behind decisions is non-transparent and challenging to decipher. To tackle this issue, the authors introduce a groundbreaking method that utilizes depth map analysis to clarify these processes, thereby making the decisions made by AI more transparent and understandable. The methodology is centred around generating depth maps from the layers of neural networks. These maps are subsequently scrutinized to gain insights into the significance of different features and the pathways leading to decisions. This strategy offers a dual advantage: it serves as a visual and a quantitative tool for interpreting AI decisions. Such an approach is pivotal in fostering AI systems that are not only more trustworthy but also reliable. The effectiveness of this method is showcased through a series of experiments, underscoring its applicability and potential benefits in various AI domains. These experiments demonstrate how depth map analysis can elucidate the internal workings of AI models, thereby making their decision-making processes more transparent and interpretable to users, researchers, and developers alike. This advancement holds significant promise for the future of AI, particularly in areas where understanding and trust in AI decision-making are crucial.

Decoding AI Decisions on Depth Map Analysis for Enhanced Interpretability

Oku,

Samayamantri,

Singhal

et al. 2024

Innovative Image Processing Methods for Colorectal Tumor Identification

Santhoshi,

Muthukumaravel

2024

Colorectal cancer is an important health issue worldwide, demanding accurate and fast diagnostic measures to reduce its impact. Traditional image processing algorithms frequently fail to effectively diagnose malignancies, resulting in incorrect diagnoses and delayed treatment. As a result, there is a critical requirement for breakthrough technologies that improve the accuracy and efficiency of colorectal tumor identification. Developed an innovative system that uses cutting-edge image processing techniques, including deep learning-based segmentation and feature extraction algorithms, to achieve more accurate and efficient tumour identification. The proposed system overcomes the limitations of existing systems by greatly increasing tumour detection accuracy, decreasing false positives and negatives, and expediting the diagnostic procedure. The proposed system consists of numerous essential parts, including data collecting, image processing, segmentation, feature extraction, classification, and model evaluation. Accurate evaluation and validation demonstrate that the proposed system consistently surpasses existing systems in terms of segmentation accuracy (93.65%), sensitivity (89.92%), specificity (94.10%), and AUC-ROC (95.63%). Furthermore, the proposed system has excellent image processing performance, including noise reduction (0.87), contrast enhancement (0.91), and image normalisation (0.93), which improves the quality and usability of colorectal tissue pictures for accurate tumour diagnosis.

Deep Neural Network for Brain Tumour Segmentation Using Guaranteed Time Slots (GTS) Algorithm

Chunduri,

Rajest,

Allugunti

et al. 2024

Segmentation of brain tumors is a crucial step in the field of medical imaging, essential for the early identification and treatment of brain tumors. This process significantly influences patient care by facilitating timely intervention. Brain tumors originate from abnormal cell growth and proliferation that diverge from normal cell behavior, resulting in tumor development. Segmentation, the differentiating between normal brain tissue and tumor regions, is critical for precise diagnosis and effective treatment strategy. Historically, the task of segmentation has been manually performed by healthcare experts. This method is labor-intensive and difficult due to the intricate and diverse brain structure. Deep neural networks (DNNs), renowned for their prowess in image classification tasks, have been employed to overcome these obstacles. Specifically, the GTS (graph-based transductive segmentation) algorithm has been introduced to optimize magnetic resonance imaging (MRI) image segmentation. This advanced method enhances the efficiency of finding the closest optimal solutions by improving convergence, thereby significantly reducing the time required for solution identification. The GTS technique stands out for its remarkable accuracy rate of 97%, showcasing some substantial improvement over previous segmentation methods. This leap in accuracy expedites the diagnostic process and increases the likelihood of successful patient outcomes by enabling earlier and more precise tumor detection.